[dbcp] Support for debugging leaking connections?

[Fredrik Jonson <fr...@myrealbox.com>]

Hello, 

I have recently been handed a project, which uses connection
poolign and appears to leak connections. Unfortunately a source
code review haven't hinted of any obvious errors.

Is there any way to make dbcp forcefully revoke connections after
some timeout? Is this built in, and if not, what is a good
approach to implementing it?

Basically, my idea is that revoking connections will sooner or
later make the application throw a nullpointer exception during
tests. 

Generally, am I approaching the problem in a sensible way, or is
there a better approach?

TIA, regards,
-- 
Fredrik Jonson

---------------------------------------------------------------------
To unsubscribe, e-mail: commons-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-user-help@jakarta.apache.org

Re: [dbcp] Support for debugging leaking connections?

[Dirk Verbeeck <di...@pandora.be>]

Hi

You are looking for the removeAbandoned feature.
http://jakarta.apache.org/commons/dbcp/configuration.html

logAbandoned will log stacktraces of the getConnection() call made to 
retrieve the lost connection. Depending on your application design 
this might be enough to find the leak.
There isn't a best approach, it all depends on your connection handling.

Connections who are allocated/freed inside a single method are easy, 
the only catch there is exception handling (close in finally block).

If you have a session or request like object you can store the 
connection in there and give it as a parameter to all the database 
methods (carefully counting these object creation/destruction, to 
debug this case).

Oracle can give you the last query executed on each connection using a 
dba tool (other database probably have this as well), this can help 
you find the error.

The last resort is to add log4j statements to each getConnection/close 
pair and analyse the logfile offline for missing close. (use NDC from 
log4j to mark each thread with a unique identifier)
You could use aspectJ for this logging but again all depends on the 
code structure of the project.

Cheers
Dirk

Fredrik Jonson wrote:
> Hello, 
> 
> I have recently been handed a project, which uses connection
> poolign and appears to leak connections. Unfortunately a source
> code review haven't hinted of any obvious errors.
> 
> Is there any way to make dbcp forcefully revoke connections after
> some timeout? Is this built in, and if not, what is a good
> approach to implementing it?
> 
> Basically, my idea is that revoking connections will sooner or
> later make the application throw a nullpointer exception during
> tests. 
> 
> Generally, am I approaching the problem in a sensible way, or is
> there a better approach?
> 
> TIA, regards,



---------------------------------------------------------------------
To unsubscribe, e-mail: commons-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-user-help@jakarta.apache.org